Confidence Intervals II 18.05 Spring 2014 Jeremy Orloff and Jonathan Bloom
Agenda Polling: estimating p in Bernoulli( p ). CLT ⇒ large sample confidence intervals for the mean. Three views of confidence intervals. Constructing a confidence interval without normality: the exact binomial confidence interval for p June 12, 2014 2 / 16
Polling: a binomial proportion confidence interval Data x 1 , . . . , x n from a Bernoulli( p ) distribution with p unknown. A ‘conservative normal’ † (1 − α ) confidence interval for p is given by z α/ 2 z α/ 2 ¯ − √ , x ¯ + √ . x 2 n 2 n Proof uses the CLT and the observation σ = p (1 − p ) ≤ 1 / 2. √ Political polls often give a margin-of-error of ± 1 / n . This rule-of-thumb corresponds to a 95% confidence interval: 1 1 x ¯ − √ , x ¯ + √ . n n † There are many types of binomial proportion confidence intervals. http://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval June 12, 2014 3 / 16
Concept question: overnight polling During the presidential election season, pollsters often do ‘overnight polls’ and report a ‘margin of error’ of about ± 5%. The number of people polled is in which of the following ranges? (a) 0 – 50 (b) 50 – 100 (c) 100 – 300 (d) 300 – 600 (e) 600 – 1000 √ Answer: 5% = 1/20. So 20 = n ⇒ n = 400. June 12, 2014 4 / 16
“...the inability to make callbacks, resulting in samples that do not adequately represent such groups as single member households, younger people, and others who are apt to be out on any given night. As overnight polls often result in findings that are less reliable than those from more carefully conducted polls, if the media reports them, it should be with great caution.” http://www.ncpp.org/?q=node/42 National Council on Public Polls: Press Release, Sept 1992 “The National Council on Public Polls expressed concern today about the current spate of overnight Presidential polls. [...] Overnight polls do a disservice to both the media and the research industry because of the considerable potential for the results to be misleading. The overnight interviewing period may well mean some methodological compromises, the most serious of which is..” ...what? June 12, 2014 5 / 16
National Council on Public Polls: Press Release, Sept 1992 “The National Council on Public Polls expressed concern today about the current spate of overnight Presidential polls. [...] Overnight polls do a disservice to both the media and the research industry because of the considerable potential for the results to be misleading. The overnight interviewing period may well mean some methodological compromises, the most serious of which is..” ...what? “...the inability to make callbacks, resulting in samples that do not adequately represent such groups as single member households, younger people, and others who are apt to be out on any given night. As overnight polls often result in findings that are less reliable than those from more carefully conducted polls, if the media reports them, it should be with great caution.” http://www.ncpp.org/?q=node/42 June 12, 2014 5 / 16
Board question A (1 − α ) confidence interval for p is given by z α/ 2 z α/ 2 √ , x ¯ + √ x ¯ − . 2 n 2 n 1 How many people would you have to poll to have a margin of 1 error of . 01 with 95% confidence? (You can do this in your head.) 2 How many people would you have to poll to have a margin of 2 2 error of . 01 with 80% confidence. (You’ll want R or a table here.) 3 If n = 900, compute the 95% and 80% confidence intervals for p . 3 June 12, 2014 6 / 16
√ answer: 1. Need 1 / n = . 01 So n = 10000. z α/ 2 2. α = . 2, so z α/ 2 = qnorm(.9) = 1 . 2816. So we need √ = . 01. This 2 n gives n = 4106. 1 1 3. 95% interval: x ± √ = x ± = x ± . 0333 30 n 1 1 4. 80% interval: x ± z . 1 √ = x ± 1 . 2816 · = x ± . 021 . 2 n 60 June 12, 2014 7 / 16
Non-normal data Suppose the data x 1 , x 2 , . . . , x n is drawn from a distribution f ( x ) that may not be normal or even parametric, but has finite mean, variance. A version of the CLT says that for large n , the sampling distribution of the studentized mean is approximately standard normal: ¯ − µ x √ ≈ N(0 , 1) s / n So for large n the (1 − α ) confidence interval for µ is approximately s s x ¯ − √ · z α/ 2 , x ¯ + √ · z α/ 2 n n where z α/ 2 is the α/ 2 critical value for N(0 , 1). This is called the large sample confidence interval . June 12, 2014 8 / 16
Review: confidence intervals for normal data Suppose the data x 1 , . . . , x n is drawn from N( µ, σ 2 ) Confidence level = 1 − α z confidence interval for the mean ( σ known) z α/ 2 · σ z α/ 2 · σ x − √ , x + √ n n t confidence interval for the mean ( σ unknown) t α/ 2 · s t α/ 2 · s √ √ x − , x + n n χ 2 confidence interval for σ 2 n − 1 2 n − 1 2 s , s c α/ 2 c 1 − α/ 2 t and χ 2 have n − 1 degrees of freedom. June 12, 2014 9 / 16
Three views of confidence intervals View 1: Define/construct CI using a standardized point statistic. View 2: Define/construct CI based on hypothesis tests. View 3: Define CI as any interval statistic satisfying a formal mathematical property. June 12, 2014 10 / 16
View 1: using a standardized point statistic Example. x 1 . . . , x n ∼ N( µ, σ 2 ), where σ is known. The standardized sample mean follows a standard normal distribution. x − µ z = √ ∼ N(0 , 1) σ/ n Therefore: x − µ √ < z α/ 2 | µ ) = 1 − α P ( − z α/ 2 < σ/ n Unwind to: σ σ P ( x − z α/ 2 · √ < µ < x + z α/ 2 · √ | µ ) = 1 − α n n This is the (1 − α ) confidence interval: σ x ± z α/ 2 · √ n Think of it as x ± error June 12, 2014 11 / 16
View 1: other standardized statistics The t and χ 2 statistics fit this paradigm as well: x − µ t = √ ∼ t ( n − 1) s / n ( n − 1) s 2 2 = ∼ χ 2 ( n − 1) X σ 2 June 12, 2014 12 / 16
View 2: using hypothesis tests Set up: Unknown parameter θ . Test statistic x . For any value θ 0 , we can run an NSHT with null hypothesis H 0 : θ = θ 0 at significance level α . Definition. Given x , the (1 − α ) confidence interval contains all θ 0 which are not rejected when they are the null hypothesis. Definition. A type 1 CI error occurs when the confidence interval does not contain the true value of θ . For a 1 − α confidence interval, the type 1 CI error rate is α . June 12, 2014 13 / 16
Board question: exact binomial confidence interval Definition. Given x , the (1 − α ) confidence interval contains all θ 0 which are not rejected when they are the null hypothesis. Use view 2 and this table of binomial(8, θ ) probabilities to: find the rejection region with significance level 0 . 10 for each 1 1 1 value of θ –put values of x with the smallest probability into the rejection region. 2 Given x = 7, find the 90% confidence interval for θ . 2 2 2 3 3 Repeat for x = 4. 3 3 2 3 2 3 2 3 3 2 3 3 3 3 3 3 θ/ 3 x 3 3 3 0 1 2 3 4 5 6 7 8 .1 0.430 0.383 0.149 0.033 0.005 0.000 0.000 0.000 0.000 .3 0.058 0.198 0.296 0.254 0.136 0.047 0.010 0.001 0.000 .5 0.004 0.031 0.109 0.219 0.273 0.219 0.109 0.031 0.004 .7 0.000 0.001 0.010 0.047 0.136 0.254 0.296 0.198 0.058 0.000 0.000 0.000 0.000 0.005 0.033 0.149 0.383 0.430 .9 June 12, 2014 14 / 16
Solution For each θ , the non-rejection region is blue, the rejection region is red. In each row, the rejection region has probability at most α = . 10. θ/ x 0 1 2 3 4 5 6 7 8 .1 0.430 0.383 0.149 0.033 0.005 0.000 0.000 0.000 0.000 .3 0.058 0.198 0.296 0.254 0.136 0.047 0.010 0.001 0.000 .5 0.004 0.031 0.109 0.219 0.273 0.219 0.109 0.031 0.004 .7 0.000 0.001 0.010 0.047 0.136 0.254 0.296 0.198 0.058 .9 0.000 0.000 0.000 0.000 0.005 0.033 0.149 0.383 0.430 For x = 7 the 90% confidence interval for p is [ . 7 , . 9]. These are the blue entries in the x = 7 column. For x = 4 the 90% confidence interval for p is [ . 3 , . 7]. June 12, 2014 15 / 16
View 3: formal Recall: An interval statistic is an interval I x computed from data x . This is a random interval because x is random. Suppose x is drawn from f ( x | θ ) with unknown parameter θ . Definition: A (1 − α ) confidence interval for θ is an interval statistic I x such that P ( I x contains θ | θ ) = 1 − α for all possible values of θ (and hence for the true value of θ ). Note: equality in this equation is often relaxed to ≥ or ≈ . = : z , t , χ 2 ≥ : rule-of-thumb and exact binomial (polling) ≈ : large sample confidence interval June 12, 2014 16 / 16
MIT OpenCourseWare http://ocw.mit.edu 18.05 Introduction to Probability and Statistics Spring 201� For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
Recommend
More recommend
